Evaluation of the software architecture styles from maintainability viewpoint

EVALUATION OF THE SOFTWARE
ARCHITECTURE STYLES FROM
MAINTAINABILITY VIEWPOINT
Gholamreza ShahMohammadi1
1

Department of Information Thechnology,
Olum Entezami University-Amin, Tehran, Iran
shah_mohammadi@yahoo.co.uk

ABSTRACT
In the process of software architecture design, different decisions are made that have systemwide impact. An important decision of design stage is the selection of a suitable software
architecture style. Lack of investigation on the quantitative impact of architecture styles on
software quality attributes is the main problem in using such styles. So, the use of architecture
styles in designing is based on the intuition of software developers. The aim of this research is
to quantify the impacts of architecture styles on software maintainability. In this study,
architecture styles are evaluated based on coupling, complexity and cohesion metrics and
ranked by analytic hierarchy process from maintainability viewpoint. The main contribution of
this paper is quantification and ranking of software architecture styles from the perspective of
maintainability quality attribute at stage of architectural style selection.

KEYWORDS
Maintainability Evaluation, Software Architecture, Architecture Style, Coupling, Complexity,
Cohesion

1. INTRODUCTION
Functionality may be achieved using a number of possible structures [1], so software architecture
styles (SASs) are selected based on the amount of their support from quality attributes. SASs
present models to solve the problem of designing the software architecture in a way that each
model describes its components, responsibilities of the components and the way they cooperate
[2]. Architectural decisions made early in the design process are a critical factor in the successful
development of the system. In particular, the selection of an appropriate architectural style has a
significant impact on various system quality attributes [3]. Since quantitative impacts of SASs on
quality attributes have not been studied so far [4], their applications are not systematic [5]. In
other words, the current use of SASs in design is ad-hoc and based on the intuition of software
developers.
A method has been shown in [6], to map an architectural style into a relational model that can be
checked for various style properties such as consistency of style. In [7], two graph-based
approaches have been shown and compared to the specification and modeling of dynamic
software architectures. The impact of a distributed software system’s architectural style on the
David C. Wyld et al. (Eds) : CST, ITCS, JSE, SIP, ARIA, DMS - 2014
pp. 183–197, 2014. © CS & IT-CSCP 2014

DOI : 10.5121/csit.2014.4118

184

Computer Science & Information Technology (CS & IT)

system’s energy consumption has been estimated in [3]. A method for specifying the relation
between six SASs and quality attributes such as maintainability has been proposed in [8]. The
relationship between the quality attributes, design principles and some SASs has been specified in
[8] using a tree-based framework. In [4], the impacts of SASs on quality attributes are determined
based on the description of style in [2]. The methods offered in [4] and [8] are not able to
determine the amount of style support from quality attributes, do not offer quantitative results
about their maintainability, and are not precise. SASs are evaluated in [9] from maintainability
viewpoint based on the scenario-based evaluation method that is less precise, less reliable and less
analyzable as compared to the measurement-based evaluation method utilized in this paper.
In [10], the performance of three SASs has been investigated through simulation-based evaluation
method. Implicit/invocation style has been verified in [11], by model checking method.
In this study, the quantitative impact of SASs on software maintainability, one of the important
quality attributes required by all software, is determined based on the measurement-based
evaluation of SASs. The SASs evaluated include Repository (PRS), Blackboard (BKB), Pipe and
Filter (P/F), Layered (LYD), Implicit/Invocation (I/I), Client/Server(C/S), Broker (BRK) and
Object-Oriented (OO), which have been introduced in [2], [12].
Software architecture evaluation methods include: 1) scenario-based evaluation, 2) simulationbased evaluation, 3) measurement-based evaluation and 4) mathematic model-based evaluation.
Measurement-based evaluation method uses metrics to measure software architecture. Metrics
evaluate internal attributes of software (e.g. coupling). External attributes (e.g. maintainability)
reflect those properties that are desirable for the software user and usually are evaluated by
internal attributes. It is believed that there is a relationship between internal and external quality
attributes. This relationship is based on theoretical models and empirical study [13], [14]. There is
a general agreement in software community that modularity has an influence on external
attributes such as maintainability [15]. Therefore, in this paper, we use coupling, complexity and
cohesion metrics to quantify the impact of SASs on software maintainability. These metrics are
essential in evaluation of software design quality and their effects on maintainability have been
extensively investigated [15]-[18].
The advantage of measurement-based evaluation as compared to scenario-based evaluation is that
the evaluation would be easier and more precise, if there are appropriate metrics. In addition, it
does not have the problems of scenario-based evaluation, namely the dependency of the results on
the scenarios used, and extensive participation of the expert. As a result, the problem is evaluated
more comprehensively.
Multi-criteria decision-making methods are used in the ranking problem of SASs. These methods
are in three categories: 1) scoring, 2) compromise and 3) concordance [19]. Analytic hierarchy
process (AHP) [20] is one of the most comprehensive multi-criteria decision making methods. It
structures the problem as a hierarchy and provides a means of decomposing the problem into a
hierarchy of sub problems that can more easily be comprehended and subjectively evaluated.
AHP reflects the way people think and behave. It also considers different quantitative and
qualitative criteria in the problem and provides sensitivity analysis on the criteria and sub-criteria.
The AHP has been proven a theoretically sound, market-tested and accepted methodology.
In this paper, to rank SASs based on the results of measurement-based evaluation of SASs, AHP
method is used.
This paper is structured as follows: Section 2 discusses software maintainability and its
measurement metrics. Section 3 explains the quantitative measurement of SASs. Section 4 deals
with the ranking of SASs. Finally, Section 5 presents the conclusion.


185

2. SOFTWARE MAINTAINABILITY AND ITS MEASUREMENT METRICS
The main objective of any software is to offer desired services according to the predetermined
quality level. There is a strong connection between many quality attributes and the software
architecture of the software system. The architecture defines the overall potential that a software
system possesses to fulfil certain quality attributes. Software are often redesigned not for the
deficiency in the functionality, but due to difficulty in maintenance, port or scale [21].
Maintainability is the capability of the software product to be modified [22]. Modifications may
include corrections, improvements or adaptations of the software to changes in the environment
and in the requirements and functional specifications. The ease of software correction is
determined through: 1) analyzability, 2) changeability, (3) stability and (4) testability [22].
A close look at software maintainability attributes reveals that provision of each characteristic
depends on the amount of modularity of software design, design with low coupling among
modules, low complexity of the modules and high cohesion of modules. Therefore, the less is the
amount of coupling and complexity of the components and the more their cohesion, the easier
will be the analyzability, changeability, stability and testability of the software. Various
researches emphasize the impact of complexity, cohesion of components and coupling among
components metrics on software maintainability [15]-[18].

2.1. Coupling Metric
High interaction of modules makes the understanding and modification of the modules more
difficult [15]. The more independent the components, the easier their understanding, modification
and maintainability [16]. Coupling is a complex concept that has been categorized by Yourdon
and Constantine [23] as: 1) Data coupling, 2) Stamp coupling, 3) Control coupling, 4) Shared
coupling and 5) Content coupling.
In this work, we generalize the “coupling among modules” concept to the coupling among
software architecture components and use it to measure the amount of coupling of SASs.
Components of SASs investigated in this work have three coupling types: data, stamp and shared
quantified based on Table 1. In [24] also consecutive numeric values from 1 to 5 were used and

the basis of such assignment was the experience from some software systems
Table 1.Type of Components Coupling
Row
1
2
3

Coupling type
Data
Stamp
Common

Symbol
w1
w2
w3

Weight
1
2
4

designs. Regarding the coupling metric, SASs are investigated in terms of the type of coupling
among the components and the number of components involved in the coupling. The more the
strength of coupling among components and the more the number of components involved in the
coupling, the less the understandability, correction and maintainability of the components [15].
To measure the coupling value of any SAS, (1) is used that is Euclidean norm, where n is the
number of style components, SCP is the amount of SAS coupling and CCPi is the amount of
coupling of the i-th component. CCPi is computed by (2), where NCTj is the number of type j
couplings, wj is the weight of the corresponding coupling type and p is the number of coupling of
the component i:

186

n

∑ CCP

SCP =

2

(1)

i

i =1
p

(2)

CCPi = ∑ NCT j .w j
j =1

2.2. Complexity Metric
Complexity value of SASs is computed by (3) where, SCM is the complexity of SAS, n is the
number of style components and CCMi is the amount of complexity of the i-th component. CCMi
is computed by (4), using the module evaluation metric of Shepperd et al [25], where fin(i) is the
fan-in of component i and fout(i) is the fan-out of component i.
n

SCM =

∑ CCM

2
i

(3)

i =1

CCMi =[fin(i)*fout(i)]2

(4)

fin(i) and fout(i) are computed by (5) and (6). In (5), Nci is the sum of the number of invocations of
component i by other components and Nri is the number of data that component i has retrieved
from the repository. In (6), Ncei is the number of other components called by component i and
Nui is the number of the repository data updated by component i. A component that controls a lot
of components usually performs various functions and so it will have a high complexity [15],[26].

fin(i)=Nci+Nri

(5)

fout(i)=Ncei+Nui

(6)

2.3. Cohesion Metric
The cohesion of a module is the extent to which its individual components are needed to perform
the same task. Types of cohesion are: 1) Coincidental, 2) Logical, 3) Temporal, 4) Procedural, 5)
Communicational, 6) Sequential and 7) Functional [23]. The cohesion type of every component is
computed based on the available information of functionality of each component of SASs
regarding the definition of the type of component cohesion in Table 2 [26].
In this work, we generalize the “modules cohesion” concept to the cohesion of software
architecture components and use it to measure the amount of cohesion of SASs. Our
investigations showed that cohesion of SASs in component are of three types: Functional,
Communicational and Logical, which are quantified based on Table 2.
Table 2. Type of Components Cohesion [26]
Cohesion type
Logical
Communicational
Functional

Description
Component performs multiple functions, and in
each calling, one of them is executed
Component refers to the same data set and/or
creates the same data set
Component performs a single well-defined
function

Symbol
C1

Weight
1

C2

2

C3

3


187

Since every component i may have different type of cohesion (Cj), so the cohesion type of
component i, CCHi, is computed by (7). Finally, cohesion of SASs is computed by (8).

CCH i = arg min C j

(7)

j
n

SCH =

∑ CCH

2
i

(8)

i =1

3. Quantitative Measurement of SASs
In this section, SASs are measured from the viewpoint of maintainability based on coupling,
complexity and cohesion metrics.
The effect of software size on SASs ranking is taken into account in the computations of this
section. In object-oriented style, the number of objects (no) and in other SASs, the number of
components (n) correspond the software size. So in the evaluations done in this section, the
number of SASs components is considered as 3, 4, 5, 6, 7, 8 and 9 and the number of classes in
object-oriented style is considered accordingly as 21, 28, 35, 42, 49, 56 and 63.

3.1. Measuring the Coupling of SASs
In this section, the coupling formula of every SAS is computed using (1) to (2).
A. Repository style. In this style, all components have common coupling with the repository.
Therefore, any change in the repository affects them. If the number of components in the
repository style is n, then the number of couplings in this style will be n as well. Thus coupling of
repository style is obtained from n . w3 .
B. Blackboard style. The control component has a common coupling with the blackboard and
has data coupling with the knowledge resources. Therefore, the control component has one
common coupling and n data coupling while the knowledge resources have a common coupling
with the blackboard. Thus, the coupling of the control component is n.w1+ w3 and the coupling
of each knowledge resource is w3. Then coupling of this style is obtained from
2
(n.w1 + w3 )2 + n.w3 .

C. Pipe and filter style Every filter (component) has a stamp coupling with the next filter while
the last filter has no coupling with any other filter. The number of couplings is n-1, and regarding
the coupling type, the coupling of this style is obtained from n − 1 w2
D. Layered style The coupling type of every layer (component) with its lower layer is data.
Considering the fact that coupling is two way, the last and first layers have only one coupling
while other layers have two couplings. So for n layer, the coupling of this style is obtained from
4 ( n − 2 ).W12 + 2 .W12 .
E. Implicit invocation style. If, in average, n/2 of components publish the events that are favored
by n/2 of the components, the coupling type of the event publisher component with the dispatcher
component is data. If an event occurs, the dispatcher component invokes the interested
components, so the coupling type of the dispatcher component with the interested components is

188


data, and the coupling of the dispatcher component will be (n/2).w1. The coupling type of
independent components (n/2) is data, so coupling of this style is obtained from
(( n / 2). w1) 2 + ( n / 2). w1 2

.

F. Client/server style. The coupling of the client with the server is data type. Supposing that, in
average, the coupling of each server component is f and, since some server components are in
transaction and usually the last component is related to repository, thus about r % of the
components have just one connection with the repository. So, coupling of this style is obtained
2
from w12 + (1 − r )n( f .w1 )2 + rnw3 .

G. Broker style. Coupling of all components is data type. Considering these facts: (1) the client
component is related to the client side proxy, (2) the client is related to the broker in order to be
informed of different services of the server, (3) the server side proxy is coupled with the broker,
(4) the broker is coupled with the server side proxy and (5) the broker is coupled with the server
for being informed of different type of services of the server, and also considering the similarity
of the coupling of the server components to client/server style, the style coupling is obtained from
2
8w12 + (1 − r )n( f .w1 ) 2 + r.n.w3 .

H. Object oriented style. In this style, the type of coupling is data. A case study done by Yu and
Ramaswamy [28] on components dependency showed that 83% of the couplings between classes
are of parameter (data) type. Coupling of each class with other classes is considered as fo. So style
coupling is obtained from f o no . 1 .

w

Column 2 of Table 3 shows the coupling formulas of SASs. The third column shows the coupling
value obtained by replacing the weight of coupling type based on Table 1.
Table 3. Coupling Formulas of SASs
Symbol
RPS

Coupling Formula

n.w
3

Coupling Value

3 n
2

2
3

(n + 3)2 + 9n

BKB

(n.w + w3) + n.w
1

P/F

n − 1 w2

LYD

4(n − 2).W12 + 2.W12

4n − 6

I/I

((n / 2).w)2 + (n / 2).w 2
1
1

(n / 2)2 + (n / 2)

C/S

2
w12 + (1 − r ) n( f .w1 ) 2 + rnw3

1+ (1− r)n. f 2 + 9.r.n

BRK

2
8w12 + (1 − r)n( f .w1 )2 + r.n.w3

8+(1−r)n. f 2 +9.r..n

OO

f o n o . w1

2 n −1

fo

no

The coupling value of classes in object-oriented style (fo) is related to the designing manner of the
past software systems. This is true for the coupling value of server components (f) in the broker
and client/server styles as well. Therefore, software designers determine the average value of
coupling (i.e. f and fo) by referring to the previous software design records. For displaying the
relationship between coupling value and software size, it is necessary that first the values of f , r
and fo parameters are determined. Thus, documents of software design projects of a large and


189

valid software company in Iran is investigated. Accordingly, after computations, the values of
these parameters become f=1.65 and fo=1.5 and r=0.2. By setting the parameters of f, fo and r to
the designated formulas and parameters n and no, the coupling value of SASs is computed
considering the software size (number of components), and its diagram is shown in figure 1.
According to this diagram, the coupling value of SASs is increased by increasing of the software
size.

Figure 1. Coupling value of SASs based on the size of software

3.2. Measuring the Complexity of SASs
In this section, the complexity formula of every SASs is computed using (3) to (6).
A. Repository style. In this style, all components read from the data repository and modify it.
Thus, both their fan-in and fan-out is equal to 1. Therefore, the total fan-out of each component,
considering the writing in the repository and invoking the repository for this writing, is 3. Thus
the complexity of independent components is 9 and the complexity of style is obtained from

9 n.
B. Blackboard style. The fan-in of the control component is 1 (for examining the status of the
blackboard) and its fan-out is 2 (for invoking the blackboard for reading its status and invoking
the knowledge resources). The fan-in of knowledge resources is 2 (for invoking by the control
component and reading from the blackboard) and its fan-out is 3 (for invocation of the blackboard
for reading and writing into the blackboard). So the complexity of the control component is 22,
the complexity of each of the knowledge resource is 36, and complexity of style is obtained
2
from 42 +36 n .

C. Pipe and filter style. The first filter (component) has no input and the last filter does not have
any output. Thus, their complexity is 0. The other filters have one input and one output. So the
complexity of style is obtained from. n −2
D. Layered style. In this style, the relation of lower layer to upper layer is response to the request
of upper layer, so in computing of layer's fan-out, this relation is ignored, i.e. only upper layer
invokes lower layer. Thus, each layer has fan-in and fan-out equal to 1. None of the layers does
not invoke first layer and the last layer invokes no layer. So their complexity is 0 and the
complexity of style is obtained from n − 2 .

190


E. Implicit invocation style. With the occurrence of an event, the dispatcher component invokes
the interested components. Therefore, the fan-in of dispatcher component is 1 (for occurrence of
the event that led to the invoking of the interested component by the dispatcher component) and
its fan-out is 1 (for invocation of the interested component, when an event occurs). Therefore, its
complexity is 1. The complexity of event publisher due to lack of fan-in and the complexity of
interested components due to lack of fan-out is 0. Therefore, the complexity of style is 1.
F. Client/server style. The client component invokes a procedure from the server, so fan-in of
the server and fan-out of the client is equal to 1. Since the client is not invoked by the components
and has no direct access to the repository, its fan-in is equal to 0 and its complexity is 0. The
number of fan-ins and fan-outs of the server components, in average, is considered as f. So the
complexity of each server component is f 4 and the complexity of style is f

4

n.

G. Broker style. The client component gets informed of the services of the server through the
method interface of the server that has been offered to the broker component, so both fan-in and
fan-out of the server becomes 1. In addition, fan-in of the broker becomes 1 due to accessing the
interface of the server services. The client invokes the client side proxy, thus its fan-out becomes
1 as well. The client side proxy sends a request to the broker component, therefore, both its fan-in
and fan-out become 1. The broker component sends the request to the server side proxy. On the
other hand, the broker invokes the server to get informed of the interface of the server services.
Therefore, both fan-in and fan-out of the broker become 2. The server side proxy has the fan-in
and fan-out equal to those of the client side proxy too. The complexity of server components is
considered similar to that of the client/server style, thus style complexity is obtained
from 274+ f 8 n .
H. Object-Oriented style. If, in average, the number of fan-in and fan-out of each class is
considered as fo, then the complexity of each class becomes fo4 and the complexity of style is

f o4 n o .
Table 4 shows the complexity formulas of SASs. Values of f and fo are considered as similar to
those in the Section 3.1.
Table 4. Complexity Formulas of SASs
Symbol
RPS

Complexity Formula

9 n

BKB

2
42 +36 n

P/F

n−2

LYD
I/I
C/S
BRK
OO

n−2
1

f4 n
274+ f 8 n

f o4 n o

By setting the parameters n and no, the complexity value of SASs is computed considering the
software size and its diagram is shown in figure 2. According to this diagram, the complexity
value of most SASs is increased by increasing of the software size.


191

3.3. Measuring the Cohesion of SASs
In this section, the cohesion formula of every SAS is computed using (7) to (8).
A. Repository style. Each component processes the same set of data, so their cohesion type is
communicational. The repository component performs various functions on the data, and in each
calling, one of the functions is performed. So its cohesion type is logical, and the cohesion of
style is n c 2 + c12 .
2

Figure 2. Complexity value of SASs based on size of software

B. Blackboard style. Each knowledge resource processes the same set of data, so their cohesion
type is communicational. The control component invokes the knowledge resources based on the
status of the blackboard. Therefore, its cohesion type is logical. The blackboard component
performs various functions and, in each invocation, one of these functions is performed. So, its
2
cohesion type is logical, and the cohesion of style is n c 2 + 2 c1 .
2
C. Pipe and filter style. Each filter processes the same set of data, so its cohesion type is
communicational and the cohesion of style is n .c 2 .
D. Layered style. Each layer contains some components; regarding the invoking of upper layer,
one of components of the lower layer is performed, so the cohesion type of each layer is logical
and the cohesion of style is n .c1 .
E. Implicit invocation style. Since the components are publisher or interested in the event, their
cohesion type is communicational. The dispatcher component performs various functions and, in
each invocation, one of them is performed. Thus, its cohesion type is logical and the cohesion of
2

2

2
style is C1 + nC .

F. Client/Server style. The server provides various services for the client by its components, and
in each invocation, one or some of the server components are performed so that each one works
on the same data. Accordingly, their cohesion type is communicational. The client component
performs a specific function, so its cohesion type is functional. The repository component

192


performs various functions and in each calling, one of them is performed. So its cohesion type is
2

2
2
logical and the cohesion of style is C 1 + nC 2 + C 3 .

G. Broker style. The client side proxy, server side proxy, broker and server components perform
various functions and in each invocation, just one of the functions is performed, so their cohesion
type is logical. The client component performs a specific function, thus its cohesion type is
functional. The repository component performs various functions and in each calling, one of them
is performed. Therefore, its cohesion type is logical. Cohesion of the server components is
considered similar to that of the client/server style. Thus, the cohesion of style

is

4C

2
1

+ nC

2
2

+

2

C .
3

H. Object-Oriented style. The classes in this style define the data of an entity and its related
functions, so, the cohesion type of each class is communicational and the cohesion of style is
no .C2
.

Column 2 of Table 5 represents the cohesion formulas of SASs. The third column shows the
cohesion value obtained by replacing the weight of cohesion type based on Table 2. By setting the
parameters n and no, the cohesion value of SASs is computed considering the software size and
its diagram is shown in figure 3. According to this diagram, the cohesion value of SASs is
increased by increasing of the software size and the amount of increase is higher in the objectoriented style relative to the other styles.

Figure 3. Cohesion value of SASs based on the software size


193

Table 5. Cohesion Formulas of SASs
Symbol

Cohesion Formula
2

Cohesion Value

2

4n + 1

RPS

n c2 + c1

BKB

n.C2 + 2C1

P/F

n . c2

2 n

LYD

n . c1

n

I/I

C1 + nC
2

C/S

C1

2

2

2

2

BRK

4C

OO

1+ 4n

2

+ nC
2

4n + 2

2

+ C3

2
2

2

10 + 4n

C

2

13 + 4n

no .C 2

1

+ nC

2

+

3

2 no

4. COMPUTATION OF THE RANK OF SASs
In this section, the ranking of SASs is performed based on the results of measurement coupling,
complexity and cohesion of SASs using AHP method.

4.1. Organizing Ranking Problem of SASs
In SASs ranking problem, aim is in the first level, metrics are in the second level and SASs are in
the third levels of the structure.

Figure 4. Hierarchical structure of SASs ranking

4.2. Computation of Priority of Metrics and the Relative Rank of SASs
In this stage, comparison matrix of the metrics and comparison matrices of SASs for the metrics
are formed. The complexity and cohesion values of a component do not affect on the other
components of SAS. However, the coupling value of a component affects the related components.
Accordingly and due to the emphasis of researches on the importance of coupling [13], [15] the
preference of coupling metric is considered more important than (1.6 ) the other metrics, and the
preferences of other metrics are considered equal. Then the relative priority of metrics is
computed by AHP method, the relative priority of coupling becomes 0.444 and that of the other
metrics become 0.278.
To determine the relative rank of SASs for each metric, comparison matrices of SASs for each
metric is formed. To set cell (i, j) of the comparison matrix of metric k, for the style x in row i
with the style y in column j, if there is a direct relation between the metric k and maintainability,

194


the ratio of the metric value of style x to the metric value of style y is set to cell (i,j), otherwise
the inverse of the ratio is set to cell(i,j). After setting of the comparison matrices based on the
described procedure, the relative rank of SASs for each metric is computed by AHP method.
Investigation of the consistency using the Expertchoice tool, tool of AHP method, showed that
consistency index is zero, so there is no inconsistency between the comparisons.

4.3. Computing the Final Rank of SASs
The final rank of SASs is computed regarding the priority of metrics and the relative ranks of
SASs. Table 6 shows the final rank of SASs. Based on the values of this Table, the Implicit/
Invocation (I/I), Pipe and Filter (P/F), and Layered (LYD) styles provide the highest support for
maintainability, respectively.
Figure 5 shows the changes in maintainability value of SASs based on the changes of software
size. With the increasing of software size, the rank of some styles such as Pipe and Filter(P/F) and
Layered (LYD) are decreased, and the rank of some styles such as Implicit Invocation(I/I) are
increased while the rank of some styles such as Blackboard(BKB) are not changed considerably.
Table 6. Rank of SASs from the maintainability viewpoint
Symbol
RPS
BKB
P/F
LYD
I/I
C/S
BRK
OO

n=3
no=21
64
52
187
185
223
95
87
107

n=4
no=28
67
54
176
169
238
97
89
110

n=5
no=35
69
54
170
161
246
97
91
112

n=6
no=42
69
55
166
155
251
98
92
114

n=7
no=49
70
55
163
151
255
99
94
115

n=8
no=56
70
55
160
149
257
99
94
116

n=9
no=63
70
55
158
146
260
99
95
116

Figure 5. Maintainability value of SASs based on the changes of software size

Figure 6 shows the diagram of styles ranks based on the relative priority of metrics. It is known as
sensitivity analysis diagram, which is drawn by Expertchoice. In this diagram, the vertical lines
show the relative priority of metrics and the horizontal lines show the rank of SASs based on the
metrics. The final rank of SASs is determined by the “OVERALL” label based on the vertical line
(figure 6). The coupling metric accords with the y-axes and after that are complexity, cohesion
and combination of the three metrics.


195

Figure 6. Diagram of styles rank regarding the relative priority of metrics

4.4. Analyzing the Rank of SAS
Here, by changing the values of some parameters, the effects of these changes on the rank of
SASs are investigated.
•

•

•

•

For the values of coupling types (Section 2.A), other values were used besides the values
mentioned in table 1 (for twelve values in the ranges 1≤w1≤1.5, 1.5≤w2≤2.5 and
2.5≤w3≤3.5), but they did not lead to any changes in the rank position of the SASs'
maintainability.
For the values of cohesion types (Section 2.C), other values were used besides the values
mentioned in table 2 (for twelve values in the ranges 1≤c1≤1.5, 1.5≤c2≤2.5 and
3≤c3≤3.5), but they did not lead to any changes in the rank position of the SASs'
maintainability.
By changing the f parameter (coupling of the server components in Section 3.A) in the
range of 1.65≤ f≤2.8 at the Client/Server (C/S) style, the change in the rank position of
this style was checked. It was found that only for f≥2, the rank position of this style is
placed after the Broker (BRK) style and no other change in the rank position of other
styles was seen.
For determining the relative priority of metrics (In Section 4.B), in addition to 1.6 (the
relative priority of coupling metric compared to that of the other metric), the ten values in
the range of 1.3 to 2.2 were used. The results showed no changes in the rank position of
the styles from maintainability viewpoint.

5. CONCLUSION
In this study, a model was offered to analyze the impact of SASs on software maintainability
according to the measurement-based evaluation of SASs. In this model, first, the formulas were
presented to compute the coupling, complexity and cohesion values of each SAS. Next, the
coupling, complexity and cohesion values of SASs were computed quantitatively using the
presented formulas. Then, the relative rank of each SAS was determined regarding the coupling,
complexity and cohesion values of SASs. Afterward, the priority of metrics was determined.
Subsequently, the final rank of SASs maintainability was determined using AHP method.
The analyses done showed that our proposed method had stability regarding the value of coupling
types, different values of f parameter, value of cohesion types and preference of coupling metric
to the other metrics.

196


Since the evaluation of this paper is based on measurement as compared to the method used in
[9], which uses scenario-based evaluation and the quality of its results is dependent on the used
scenarios and also on the extensive expert participation, the results of our proposed model is more
precise, more reliable and more analyzable.
The proposed method gives formulas to determine the values of 1) coupling, 2) complexity and 3)
cohesion of each SAS, while this has not been done in previous methods.
As compared to [4], [8], both the proposed method and the method used in [9] give the
quantitative results about the maintainability of SASs that is basis of the systematic
recommendation and selection of SAS.
Finally, only the proposed method examines the effect of the software size on the maintainability
rank of SASs.
The methods given [6], [7], [11] use the mathematic model-based evaluation and the method used
in [10] uses the simulation-based evaluation. These methods verify specific features such as
consistency and satisfaction of some properties by SASs that are different from the quality
attributes required in this paper. The above points and table 8 clearly show the position of the
proposed method as compared to the methods of [4], [8] and [9].
It is worth noting that the ranking of SASs based on our proposed method is consistent with the
priorities of SASs from the viewpoint of maintainability in the methods used in [2], [12], which
are based on experimental studies.
Table 7. Comparison of the proposed method with the related methods
Method

Proposed
Method

Criteria
Base
Offering the Quantitative Results
about the Maintainability of SASs
Total SASs that were Investigated
Considering the Effect of Software
Size on the Rank of SASs

Measurement
●
8
●

Method
[4]
Tree

Method
[8]

Method [9]

Unsystematic

Scenario
●

6

8

REFERENCES
[1]
[2]
[3]

[4]
[5]
[6]
[7]

Len Bass, Paul Clements. & Rick Kazman(2003) Software Architecture in Practice (2nd Edition),
Addison-Wesley, p 89.
F. Buschmann, R. Meunier, H. Rohnert, P. Sornmerlad, & M. Stal,(1996) Pattern-Oriented Software
Architecture- A system of Patterns" John Wiley & Sons, p. 394.
C. Seo, G. Edwards, S. Malek, & N. Medvidovic,(2009) "A Framework for Estimating the Impact of
a Distributed Software System’s Architectural Style on Its Energy Consumption", 7th Working
IEEE/IFIP Conf. on Software Architecture, pp. 277-280.
B. Harrison, & P. Avgeriou,(2007) "Leveraging Architecture Patterns to Satisfy Quality Attributes",
1st European Conf. on Software Architecture, Springer, pp. 263-270.
P. Avgeriou P, & U. Zdun, (2005) "Architectural Patterns Revisited: A Pattern Language", Proc. of
10th European Conf. on Pattern Languages of Programs, pp.1-39.
J.S Kim, and D. Garlan, (2006) "Analyzing Architectural Styles with alloy", Proc. of the ISSTA 2006
workshop on Role of Software Architecture for Testing and Analysis, pp. 70-80.
R. Bruni, A. Bucchiarone, A. Gnesi, D. Hirsch, & A.L. Lafuente, (2008) "Graph-based Design and
Analysis of Dynamic Software Architectures", LNCS 5065, pp. 37–56,.

[8]
[9]

[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]

[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]

197

H. Reza, & E. Grant, (2005) "Quality-Oriented Software Architecture", the IEEE Int. Conf on
Information Technology, pp. 140 – 145.
Gholamreza Shahmohammadi, & Saeed Jalili, (2009) "Scenario-Based Quantitative Evaluation of
Software Architecture Style from Maintainability Viewpoint", 14 th Annual of CSI Computer
Conference (CSICC 2009), Iran, Amirkabir University.
H. Grahn, & J. Bosch, (1998) "Some Initial Performance Characteristics of Three Architectural
Styles", Proc. of Int. Workshop on Software and Performance.
D. Garlan, & S. Khersonsky, (2000) "Model Checking Implicit Invocation Systems", 10th Int.
Workshop On Software Specification and Design.
M. Shaw & D. Garlan, (1996) Software Architecture: Perspectives Discipline on an Emerging
Discipline, Prentice Hall.
L. Briand, S. Morasca, & V. Basili, (1996) "Property Based Software Engineering Measurement",
IEEE Trans on Software Eng., vol. 22, no. 1, pp. 68-86.
L. Briand, J. Wust, & H. Lounis, (1999) "Using Coupling Measurement for Impact Analysis in
Object-Oriented Systems", IEEE Int. Conf. on Software Maintenance.
S.L. Pfleeger, & J.M. Atlee, (2006) ”Software Engineering, Theory and Practice”, 3rd Edition,
Prentice Hall.
P. Yu, T. Systa, & H. Muller, (2002) "Predicting Fault Proneness using OO Metrics. An Industrial
Case Study," 6th European Conf. on Software Maintenance and Reengineering, pp.99 – 107.
M. Alshayeb, and L. Wei, (2003) "An Empirical Validation of Object-Oriented Metrics in Two
Different Iterative Software Processes," IEEE Trans on Software Engineering, vol. 29 (11), pp. 1043
– 1049.
F. Bachmann, L. Bass, M. Klein, M. & C. Shelton, (2005) “Designing Software Architectures to
Achieve Quality Attribute Requirements”, IEE Proc. of Software, Vol. 152, No 4,pp. 153- 165.
C.L. Hwang, K. Yoon, (1981) "Multiple Attribute-Decision Making", Springer-Verlag.
T. L. Saaty, & L. G. Vargas, (2001) “Models, Methods, Concepts & Applications of the Analytic
Hierarchy Process”, Kluwer Academic Publisher.
L. Bass, P. Clements, & R. Kazman, (1998) Software Architecture in Practice, Addison-Wesley, p.
17.
ISO, (2001), International Organization for Standardization, “ISO 9126-1:2001, Software Engineering
– Product quality, Part 1: Quality model”.
E. Yourdon, & L. Constantine, (1978) Structured Design, Englewood Cliff, NJ, prentice Hall.
N. Fenton, & A. Melton, (1990) "Deriving Structurally Based Software Measures", Journal of
Systems and Software 12(3), pp. 177-187.
M. J. Shepperd, & D.C. Ince, (1990) "The use of metrics in the early detection of design errors",
Proc.of the European Software Engineering Conf, pp.67-85.
NE. Fenton, & SL. Pfleeger, (1997) "Software Metrics: A Rigorous and Practical Approach", (2nd
Edition), International Thomson Computer PRESS.
S. Chidamber, & C. Kemerer, (1994) "A Metrics Suite for Object Oriented Design", IEEE Trans on
Software Engineering, vol. 20, pp. 476-493.
L. Yu, & S. Ramaswamy,(2007) "Component Dependency in Object-Oriented Software", Journal of
Computer Science and Technology, 22(3), pp. 379-386.
Lee, S.hyun. & Kim Mi Na, (2008) “This is my paper”, ABC Transactions on ECE, Vol. 10, No. 5,
pp120-122.

AUTHORS
Gholamreza Shahmohammadi received his Ph.D. degree from Tarbiat Modares
University (TMU, Tehran, Iran) in 2009 and his M.Sc. degree in Computer
Engineering from TMU in 2001. Since 2010, he has been Assistant Professor at the
Olum Entezami University-Amin(Tehran, Iran). His main research interests are
software engineering, quantitative evaluation of software architecture, software metrics
and software cost estimation.
E-mail: Shah_mohammadi@yahoo.co.uk

Evaluation of the software architecture styles from maintainability viewpoint

More Related Content

What's hot

Viewers also liked

Similar to Evaluation of the software architecture styles from maintainability viewpoint

Recently uploaded

Evaluation of the software architecture styles from maintainability viewpoint